Moxi 2.0 Tests the Value of Robotic Fleet Data as a Commercial Differentiator in Healthcare Settings

On August 17, 2026, Diligent Robotics began its deployment of Moxi 2.0, the second generation of its hospital delivery robot at Endeavor Health Edwards Hospital, Providence Saint John’s Medical Center, and Children’s Hospital Los Angeles (CHLA) on the back of existing Moxi 1.0 fleets in more than 25 facilities. Moxi is a mobile manipulator, a wheeled robot that navigates a building independently and carries a single arm suited for pressing elevator buttons and opening its integrated drawers as it moves supplies, medication, and samples across departments. Serve Robotics, an early pioneer of Autonomous Mobile Robots (AMRs) intended for sidewalk delivery, completed its acquisition of Diligent in January 2026, diversifying the Original Equipment Manufacturers’ (OEMs) portfolio in greenfield robotics markets.

Moxi 2.0’s key innovation is Diligent’s proprietary, nosocomial world model, which uses fleet deployment data to improve short-horizon perception, prediction, and behavioral adaptation in crowded clinical environments. That equips it with the necessary tools to navigate busy hospital corridors, where gurneys, clinicians, and large devices engage in haphazard, unstructured movement within a public setting. Relative to the first-generation Moxi, Diligent claims Moxi 2.0 offers 10X onboard compute, 10X to 15X faster perception, and up to 18 hours of operating time a day, with recent Serve disclosures pointing to NVIDIA RTX A2000-based compute and optimized charging.

Moxi 2.0 runs on a data flywheel, meaning each deployment expands the operational dataset that Diligent can use to refine its world model across the fleet. Deployment therefore becomes part of development: more units in more hospitals should generate more edge-case data, improving model performance and making Moxi easier to justify operationally. Moxi’s deliveries cluster around the same repetitive cases (passing narcotics, samples, supplies, etc. from department to unit), producing high-density data that can refine the robot’s performance of routine tasks. In healthcare robotics, where many OEMs remain trapped in pilot purgatory, deployment can become an operational asset as opposed to a mere sales milestone.

What separates Diligent from other OEMs is mobile manipulative capacity. Aethon, a pioneer in robotic solutions for healthcare logistics, has already succeeded in commercially deploying a suite of AMRs. Its T3 and Zena RX models autonomously navigate hospital floors and reach myriad departments by integrating with elevators, doors, and fire systems, and have been accumulating navigation data across installments for over a decade. Those fleets, however, carry payloads in enclosed compartments and rely entirely on staff to load and unload them, which makes Diligent’s syncretism of autonomous mobility and dexterous manipulation especially valuable in high-risk, greenfield markets.

A competitor can buy the same NVIDIA compute, commission a comparable arm, and attach it to an AMR base within a short development period. Still, it cannot purchase years of Moxi’s hospital corridor experience. In coming quarters, it will be imperative to identify performance gains directly attributable to Diligent’s world model, reinforcing the value of fleet data and flywheel machine learning in physical Artificial Intelligence (AI).

Serve’s acquisition of Diligent itself illuminates the perceived value of fleet data in robot development. Having built its business on sidewalk delivery solutions, already among the riskier use cases for automation given uncontrolled pedestrian environments and public liability, Serve moved into a harder one still—mobile manipulation inside occupied clinical space, under chain-of-custody requirements. Rather than diversifying toward a lower-risk market post-merger, Serve concentrated its exposure within complex automation, betting that the data flywheel itself will be a transferable, growth-oriented asset.

Regulation explains much of why healthcare has automated so little, at least in the United States, even as pilots continue to attract attention and investment. The world’s most widely deployed surgical robotics solution, Intuitive Surgical’s da Vinci, still relies largely on teleoperation. Higher autonomy would trigger a more demanding Food and Drug Administration (FDA) review pathway, potentially including De Novo classification or, if risk cannot be mitigated, stringent and costly Class III premarket approval. Worker safety creates a bottleneck alongside FDA consideration, as the Occupational Safety and Health Administration (OSHA) plainly states that there are no specific regulations for the robotics industry, leaving hospitals to build safety policy from general machine guarding rules and voluntary consensus standards, which are most often produced with an industrial focus that is too broad for clinical applications. Diligent’s advantage is an ability to expand automation within the narrow regulatory and operational gap hospitals will currently accept.

Fleet learning methods are not new to robotics. Warehousing, logistics, and manufacturing automation programs have refined deployment data models for several years, and similar approaches (such as simulation-to-real training and robot gyms) are spreading to new Physical AI categories. Its arrival in clinical mobile manipulation is unique, as this setting has resisted automation due to unstructured tasks, deformable objects, and low tolerance for error. The Moxi 2.0 deployment signals that mobile manipulation has started to leech into greenfield clinical markets. However, long-term success is not guaranteed, and the model is buoyed by strong publicity.

• Many robotics pilots represent sunk costs, and, when unsuccessful, can create reputational drag, dampening prospects for further automation both within the same facility and across peer institutions.
• Density carries higher importance than breadth. For learning flywheel systems, a few sites running continuously will produce a matured, more refined model than would a wide scatter of low-utilization pilots, though actualizing that kind of deployment involves heavy financial costs and stringent regulatory hoops––CHLA has logged more than 40,000 Moxi deliveries but has only grown from two units to three.
• Manipulation data and navigation data are not one in the same. Mobile robot OEMs claiming a data advantage should be able to delineate what share of their portfolio involves advanced capabilities beyond movement, such as dexterous manipulation, which is the real point of ingenuity in hospital logistics.

Perhaps the biggest question lies not within Moxi’s impressive capabilities, but whether the announcement of its advancements is simply a narrative effort to offset deteriorating investor sentiment. Serve Robotics announced the Moxi 2.0 deployment on August 17, just 11 days after cutting full-year revenue guidance from roughly US$26 million to a range of US$9 million to US$10 million. The market did not reward Serve’s announcement, with shares falling 7.1% the next day. Two disclosures on Diligent’s behalf can settle this contention: an intervention rate showing how often staff must step in per thousand Moxi deliveries, and the time required to commission Moxi’s use within new facilities. Serve Robotics’ next quarterly results will suggest whether the model is deploying fast enough—and accurately enough—to keep its own flywheel turning.